Prediction of the pH effect on arsenic (V) removal by varying catalyst of magnetic xerogel monoliths based on FREN model

Contamination of groundwater by arsenic is still a widespread world problem and directly affects millions of people in many countries such as the USA, Argentina, Australia, Bangladesh, Chile, China, India, Mexico, Taiwan, Thailand and Vietnam (Pal et al. 2009). The risk of human exposure to inorganic arsenic through drinking water can lead to significant health issues (Shankar et al. 2014). Arsenic can cause cardiovascular, hepatic, neurological and dermal diseases, problems in the immune and endocrine system, diabetes, cancer (skin, liver, lung and bladder), gastrointestinal absorption, skin and respiratory tract (Huang et al. 2015). Adsorption has been used extensively in water treatment due to high efficiency, easy operation and handling, and variety of adsorbent materials that can be regenerated and reused (Bonilla-Petriciolet et al. 2017).

The resorcinol-formaldehyde gels (RF) are mesoporous materials with properties of texture, mechanical resistance and chemical stability. The variation of the synthesis and processing conditions can be controlled and altered to their final nanostructures of the RF gel (Al-Muhtaseb & Ritter 2003). Doping with metal and metal oxides is interesting, especially applications in the field of adsorption and catalysis (Maldonado-Hódar et al. 2004; Minovic-Arsic et al. 2016; Ribeiro et al. 2016). The application of magnetic adsorbents for water and wastewater treatment are widely used, especially iron oxides due to their abundance, easy magnetically separation, affordability, and applicability in broad pH range (Mehta et al. 2015).

This research focused on the ability to synthesize and generate magnetic adsorbents through the sol-gel method. Indirect sonication methods were applied to fix the magnetite particle in RF aqueous solution before the gelation and curing process because formaldehyde aids metal ions (Fe³⁺) to precipitate under metallic state and the effects on non-homogeneous and metal agglomerates into the gel (Job et al. 2004).

The materials synthesized by varying molar ratio (R/C) of resorcinol and catalyst were characterized by means of analytical techniques including pH at the point of zero charge (pH_PZC), isoelectric point (IEP), scanning electron microscopy (SEM), energy dispersive X-ray spectroscopy (EDX), X-ray diffraction (XRD), nitrogen adsorption and Fourier transform infrared spectroscopy (FTIR). A batch adsorption experiment was used to investigate these materials with As(V) 0.1 g/L of groundwater as a function of solution pH (2–7). The varying R/C ratios of the resultant magnetic adsorbents, the effect on their characterizations, and their capacity for arsenic adsorptions, were evaluated. The prediction was also applied using an adaptive network called fuzzy rules emulated network (FREN) (Treesatayapun & Uatrongjit 2005; Khamkure et al. 2013; Treesatayapun 2017). By taking advantage of human knowledge according to the arsenic removal and pH property, the network architecture (Treesatayapun 2016) of FREN was established altogether with the data base system developed by the data of experimental results. In this work, the database systems were established according to the human knowledge of arsenic removable for the materials. Furthermore, the learning algorithm was developed for accumulating the sample data. By using the prediction model based on FREN, the performance of tested materials can be evaluated according the desired pH range.

Resorcinol (1,3-dihydroxybenzene C₆H₄(OH)₂, 99.21%, Meyer), formaldehyde (HCHO, 37% methanol stabilized solution, J. T. Baker), sodium carbonate (Na₂CO₃, granular anhydrous, J. T. Baker), acetone ((CH₃)₂CO, 99.5%, J. T. Baker) and magnetite (Fe₃O₄, Lanxess) were used to synthesize magnetic adsorbents. Ultrapure (type1) water obtained by water purification system (Labconco, Waterpro BT) was used in the synthesis and was used to prepare all the solutions in the study.

Sodium arsenate dibasic heptahydrate (HAsNa₂O₄.7H₂O, Sigma-Aldrich) was used to prepare As(V) stock solution by dissolving to a concentration of 1,000 mg/L. As(V) solutions used for studying pH effect in the batch experiment was obtained by diluting the As(V) solution to a concentration of 0.1 mg/L with groundwater.

Magnetic xerogels (MCs) were synthesized from the polymerization of resorcinol (R) and formaldehyde (F) in deionized water (W), using sodium carbonate as the catalyst (C) following the procedure described by Pekala (1989) and finally mixing with magnetite (M). Molar ratios used in this synthesis were: R/F = 0.5, R/W = 0.06, M/R = 0.01 (Haro et al. 2011; Ribeiro et al. 2016) with varied molar R/C in the values 50, 100 and 200. These are labelled as MC50, MC100 and MC200, respectively.

The mixture was stirred in magnetic agitation at room temperature and homogenized for 15 min. Magnetite (M) was added into homogeneous RF aqueous solution and then ultrasonicated by using an ultrasonic processor (model UP400St; Hielscher) to disperse magnetite particles. After indirect sonication for 60 min, MCs were produced by visual determination of magnetite distribution into xerogel. MCs were placed in an oven set at 80 °C for 5 days to ensure complete crosslinking in the curing process. The gels were then removed from the conventional oven and allowed to cool at room temperature. After curing, samples were removed from the test tubes and cut into pellets approximately 5 mm in length using a diamond disk. In order to remove water from their structure, the solvent exchange was applied so the gels were immersed in acetone for 2 days at room temperature by use of an agitator (model; BS-11; Lab companion). After vacuum filtration, solvent was replaced daily. Finally, the gels were dried in an oven at 80 °C for 3 days. The dry materials obtained were black-colored polymers and labelled MCs. The incorporation of magnetite in the structure of RF gel and the effect of modified ratio of catalyst were studied.

Surface chemistry of materials was obtained by pH at the point of zero charge (pH_PZC). Zeta potential analyzer (Microtrac, PMX 500) was used to measure the zeta potential with a pH range (2–11) to determine the isoelectric point (IEP). The morphology and the punctual chemical composition of the materials were analyzed by field emission scanning electron microscopes. FE-SEM and energy dispersive EDX, using a JEOL JSM-7800F Prime, which were coated with gold. XRD was performed by Xpert Phillips PW3040. The measurement of surface area and pore size of MCs were determined by the analysis of nitrogen adsorption isotherm by using analyzer model ASAP 2020, Micromeritics. Prior to analysis, the condition of degasification of MCs were dried at 110 °C for 12 h. The textural properties were carried out by N₂ at 77 K. The Brunauer–Emmett–Teller (BET) surface area was determined from N₂ adsorption-desorption isotherms by using BET equation form relative pressure (P/P₀) between 0.1 and 0.3. The total pore volume was calculated from the amount of N₂ adsorbed at P/P₀ = 0.975. The mesopore and micropore volume was obtained by using Barret–Joyner–Halenda (BJH) and density functional theory (DFT) methods, respectively (Morales-Torres et al. 2010; Verma et al. 2015). The pore size distribution was estimated by using DFT method. Fourier transform infrared (FTIR) was carried out through a spectrophotometer (Shimadzu, model IRAffinity-1S) to determine the surface functional groups that are located in the synthesized organic xerogels. The FTIR analysis of the three samples was carried out before and after exposure of As. The samples in powder form were dried overnight at 60 °C before placement in a desiccator awaiting analysis. The infrared spectra results were acquired between the wave number of 400–4,000 cm⁻¹ with 45 scans per sample. Attenuated total reflection (ATR, Specac; GS10800) with the crystal puck (type IIIa monocrystalline diamond) was installed and fitted to the optical unit of FTIR system

The effect of pH in aqueous solution plays an important role in the process of arsenic adsorption due to its effects on the protonation and deprotonation of the adsorbent. Normally, natural water has a pH in the range of 5.0–8.0 and the predominant arsenate species are and (Pal et al. 2009).

The groundwater sample used in this experiment was obtained from a water well located in Jiutepec, Morelos, Mexico. It was analyzed according to a standard method. Physical and chemical properties of the studied groundwater are shown in Table 1. Several parameters (pH, TDS, Cl⁻, F⁻, As, Fe, Mn, and ⁠) fulfilled the Mexican regulations (NOM-127-SSA1-1997).

The adsorption tests were carried out individually using solution containing 0.1 mg/L of As(V) and was studied as a function of pH. The initial pH of the solutions was adjusted from 2 to 7 by using 0.1 M HCl and 0.1 M NaOH solutions. All batch adsorption experiments used 50 mg (dosing 1 g/L) of adsorbent added in the centrifuge tube with 50 mL of As(V) solution. The plastic tubes were next shaken using agitation equipment (Companion; model BS-11) at 150 rpm for 6 h at room temperature (25 ± 2 °C). Afterwards, the solutions were filtrated through vacuum pump filtration using a Whatman glass microfiber filter grade GF/A to separate the solid from the supernatant. Determination of the residual concentration of As(V) was completed by using arsenator digital arsenic test kit (PT981, Palintest) with the reading value of 10 μg/L As(V) in drinking water in regard to WHO guideline.

According to the experimental results from batch adsorption of As(V) as a function of pH, the performance evaluation system is utilized for those raw data. The parameter ‘qe’ denotes the removal index of As(V) from the treated water. In this work, the proposed evaluation system is established by using the variation of pH with respect to qe regarding the material MC50, MC100 and MC200.

The block diagram in Figure 1 represents the overall system for the evaluation scheme. Two adaptive networks called fuzzy rules emulated network (FREN) are implemented. The first network is FRENq which is described in Figure 2(a). This network has two inputs as pH value and the varying ratio of R/C which denotes the type of materials. The output is the estimated qe which is stored in the database system as the block diagram in Figure 1. The database system has the content as the human knowledge which is related to the property of the materials. The data transformation between the database system and the storage unit enhance the connectivity for the next stage.

In the next stage, another network named FRENp is developed to determine the performance index ‘pI’ of the material with respect to qe parameters obtained by FRENq. The parameter pI is established by the basic knowledge for the human sense as the following IF-THEN rule: IF qe is high (Good As(V)-removal) THEN pI should be high value.

For both FRENq and FRENp (Figure 2), the internal parameters named linear consequence (LC) are automatically tuned by the conventional gradient search. Furthermore, FRENq is also enhanced by the R/C ratio delivered from the database system.

The example of LC parameters is illustrated in Figure 3(a). Furthermore, the membership function (MF) parameters are also given as the plot in Figure 3(b).

The leaning phase is stopped when the LC parameters are all steady. This can prevent the overfit problem. The learning rate can be given as a small constant, and it can be increased for improving the convergence speed.

IEP is the point where the MCs exhibit no net charge. The IEP of the MC50, MC100 and MC200 are 3.4, 3.59 and 3.7, respectively. pH_PZC is pH at which the surface of adsorbent is neutral. The pH_PZC of the MC50, MC100 and MC200 are 6.63, 6.12 and 4.35, respectively. Therefore, the adsorbent surface of MC50 is positively charged at pH < 6.63 and becomes negatively charged at pH > 6.63 (Thanh et al. 2019).

The morphology and microstructure of MCs were characterized by using SEM as shown in Figure 4. It can be observed that the SEM images of MCs are composed of large numbers of microclusters with a three-dimensional network and can present as a porous structure of materials (Oyedoh et al. 2013; Prostredný et al. 2018). The distribution of magnetite that could be observed was in uniform distribution (Minovic-Arsic et al. 2016). Figure 4(a) shows MC50 has more compact micro clusters than MC100 (Figure 4(b)). Figure 4(d) indicates MC200 is less compact with micro clusters than MC100 (Figure 4(c)). Thus, it can be concluded that the size of micro clusters is increasing following the R/C ratio of MCs increase. These results are in agreement with results reported by Oyedoh et al. (2013), indicating that low R/C ratio has an effect on the smaller size of micro clusters. Conversely, high R/C ratio makes a difference in having the bigger micro cluster size. Some parts of the surface morphology of MCs can be observed to have the agglomeration of unbound Fe contents into the RF matrix. It can be conjectured that during the synthesis, the magnetite doping into RF gel occurred before the curing step and did not participate with the polycondensation reaction between resorcinol and formaldehyde. Thus, it can be shown as a physical dispersion of the Fe contents into RF gel, similar to the results obtained by Verma et al. (2015). The EDX spectra of MCs indicated the RF gel was composed of the elements C, O and Fe. The presence of Fe content can be confirmed to be approximately 1.19% within the RF gel which is a similar result to the theoretical calculation (approximately 1.47%). In the process of solvent exchange with acetone, it can be observed that some magnetite particles leached out from the MCs at first because their particles were unbound from the polymetric matrix during synthesis.

Figure 5 represents XRD patterns displaying diffraction peaks of MCs with different R/C ratios. All materials have similar patterns. The shape of each peak was broadened due to the influence of RF polymers that have similar patterns with halo peak at 2θ = 15–20° according to the results of Oyedoh et al. (2013). The diffraction peaks of MCs at 2θ values close to 10° correspond to the results of Huang et al. (2018). This indicates these are typical of mesoporous materials.

Their XRD results with diffraction peaks at 2θ values of 18°, 30°, 35.5°, 43°, 57° and 62° correspond to the crystallographic planes of magnetite 111, 220, 311, 400, 511 and 440, and respond to ICCD card number 00-01900629 similarly reported by Tipsawat et al. (2018). The position and relative intensity of all diffraction peaks of samples were consistent with the crystalline pattern of Fe₃O₄ phase. Therefore, the addition of magnetite into the synthesis of xerogel did not cause any phase changes.

The results of textural characteristics of MCs with varying molar ratio of R/C were analyzed. The BET surface area of MC50, MC100 and MC200 are 365.9, 545.1 and 529.5 m²/g, respectively. Total pore volume is between 0.255 and 0.683 cm³/g with the average pore size in the range of 2.787–5.160 nm. It seems that BET surface area, total pore volume and average pore diameter of MCs increases with the increasing molar ratio of R/C. The use of a higher amount of catalyst during synthesis effectively lowered surface area and porosity. The percentage of volume of mesopore of the MCs increased proportionally with R/C ratio. This conclusion was confirmed by the observation in the SEM images and the diffraction peaks of MCs at 2θ values close to 10° shown in Figures 4 and 5, respectively. Thus, it can be summarized that the surface morphology of RF gel is directly affected by the variation of catalyst (Mirzaeian & Hall 2009; Prostredný et al. 2018).

Adsorption isotherm of N₂ on the MCs obtained at different R/C ratio is given in Figure 6(a). MC50, MC100 and MC200 exhibit N₂ adsorption isotherm of type IV, which is the typical of mesoporous materials (Morales-Torres et al. 2010; Morales-Torres et al. 2012). It is similar to the results obtained by Bekyarova & Kaneko (2000) at pH = 7, reporting that the initial pH solution and the nature of the dopant metal affect the sol-gel chemistry and the structure. The initial pH and pH after mixing with magnetite of aqueous solution were in the range of 6.87–7.37 and 6.82–7.26, respectively. This was in the pH range reported by Pekala (1989). The pH of the solution after the addition of Fe₃O₄ to MC50, MC100 and MC200 were 7.26, 7.05 and 6.82, respectively. pH of solution decreased. The results of both surface area and pore volume increased.

The results of N₂ adsorption isotherms can be explained that with the increase of S_BET, total pore volume increase and volume of mesopore of MCs increase when molar of R/C increases. The comparison of pore size distribution (PSD) of magnetic xerogels with varying molar ratio of R/C (Figure 6(b)) demonstrates that the increase of R/C ratio leads to the increase in mesoporosity. A higher amount of catalyst (low molar ratio of R/C) leads to lowering of the surface area including pore size and pore volume decrease. These results are in agreement with results reported by Oyedoh et al. (2013) and Job et al. (2004) indicating that larger R/C ratio leads to have larger polymer modules of MCs and during wet gel produces larger pore size. After drying the MCs, the surface tension is decreased but the pore size is larger. This leads to a low shrinkage with conservation of the larger pore size and pore volume.

Figure 7(a) and 7(b) shows the FTIR studies of functional groups presented in MCs (produced by a varying R/C molar ratio) before and after arsenic adsorption, respectively. The resultant infrared spectra were acquired between the wave number of 600–4,000 cm⁻¹ with 45 scans per sample. The results show the major of six adsorption bands at 3,300 cm⁻¹ (-OH group), 2,900 cm⁻¹ (C-H stretching), 1,600 cm⁻¹ (C = C aromatic stretching), 1,400 cm⁻¹ (C-H bending vibration), 1,200 cm⁻¹ (C-O stretching), and 457 cm⁻¹ (Fe-O stretching vibration), respectively. The obtained spectra peaks agreed with those of similar reported researches (Morales-Torres et al. 2010; Verma et al. 2015). After As adsorption, it can be observed that the presence of the new band peak with small intensity at 1,286–1,288 cm⁻¹ is due to As-OH bond stretching (Gore et al. 2018). Meanwhile, the band at 1,090 cm⁻¹ corresponds to metal hydroxyl (M-OH) stretching vibration decreasing considerably after the adsorption, similar to the results obtained by Min et al. (2017). This confirms the adsorption of As ions on MCs.

The initial pH of the solution is the important factor for the adsorption process. The pHpzc values of the prepared MCs were found in the range of 4.35–6.63. If pH is below the pHpzc value, the surface charge of adsorbent would be positive, hence the anions can be adsorbed. When pH is over the pHpzc value, the surface of MCs becomes negatively charged and incompatible for adsorbing anionic components. In this study, the adsorptive capacity or the uptake of As(V) on MCs (qe), as a function of pH (2–7) was investigated. The uptake of arsenic on MCs (qe), represented in Figure 8(a), were varied at different pH. The maximum qe was obtained by MC50 with 72 μg/g (73.5% As(V) removal) at pH = 5. qe of MC50 decreased at pH 3, 4 and 7. In contrast, qe of MC100 seemed to work well at pH 3, 6 and 7. The adsorption of As(V) onto MC100 reached the maximum for both weakly acidic conditions (pH = 3) and the neutral region (pH = 7). It can be explained that the predominant arsenate species are and adsorbed on the MCs by substituting hydroxyl ions (Min et al. 2017). It is confirmed that the As(V) adsorption of MCs in groundwater strongly depended on the pH under the experimental conditions. This result is similar to that obtained by Minovic-Arsic et al. (2016) for As(III) adsorption. The highest adsorption capacity of MCs will be selected as an optimum pH condition for further experiments of the effect on operating parameters, adsorption of kinetics modeling, isotherm, and thermodynamics.

The qe estimation results are depicted in Figure 8(b) for MC50, MC100 and MC200. It is clear that FRENq has the ability to estimate the arsenic removal along the range of pH. Furthermore, the results of performance evaluation as presented as pI value mentioned above of MC50, MC100 and MC200 are 195.8767, 182.331 and 78.4791, respectively. According to these results, the pH range between 2 and 7, MC50 has better performance. However, for treating water with a pH between 6 and 7, MC100 has higher performance index.

Magnetic xerogel can be simply synthesized with the sonication of magnetite into the mixed RF solution. The incorporation of magnetite in the structure of RF gel was physical without participation in the polycondensation reaction of RF corresponding to XRD patterns and SEM figures. The influence of surface morphology and textural properties of MCs is directly affected by the varying of the catalyst. MCs provide the ability for As(V) removal confirmed by FTIR spectra after the adsorption. The performance of FREN prediction has been evaluated using the experimental data. According to the results, MC50 presents the best performance for the wide range of pH while MC100 shows the best performance within the application range of pH. The predictive scheme can be implemented to evaluate the arsenic removal performance by using the limited data from experimental results. Further research on MCs for water treatment should investigate the magnetite loading, drying procedures (super- or subcritical drying) and carbonization of RF xerogel to improve the adsorption capacity.

This work was supported by Consejo Nacional de Ciencia y Tecnología (CONACYT), Cátedras 2017-159.

All relevant data are included in the paper or its Supplementary Information.